Search

Neighborhood regularization is highly important for a latent factor (LF)-based Quality-of-Service (QoS)-predictor since similar users usually experience similar QoS when invoking similar services. Current neighborhood-regularized LF models rely prior information on neighborhood obtained from common raw QoS data or geographical information. The former suffers from low prediction accuracy due to the difficulty of constructing the neighborhood based on incomplete QoS data, while the latter requires additional geographical information that is usually difficult to collect considering information security, identity privacy, and commercial interests in real-world scenarios. To address the above issues, this work proposes a posterior-neighborhood-regularized LF (PLF) model for QoS prediction. The main idea is to decompose the LF analysis process into three phases: a) primal LF extraction, where the LFs are extracted to represent involved users/services based on known QoS data, b) posterior-neighborhood construction, where the neighborhood of each user/service is achieved based on similarities between their primal LF vectors, and c) posterior-neighborhood-regularized LF analysis, where the objective function is regularized by both the posterior-neighborhood of users/services and L2-norm of desired LFs. Experimental results from large scale QoS datasets demonstrate that PLF outperforms state-of-the-art models in terms of both accuracy and efficiency.

Relation extraction between entity pairs is an increasingly critical area in natural language processing. Recently, the pre‐trained bidirectional encoder representation from transformer (BERT) performs excellently on the text classification or sequence labelling tasks. Here, the high‐level syntactic features that consider the dependency between each word and the target entities into the pre‐trained language models are incorporated. Our model also utilizes the intermediate layers of BERT to acquire different levels of semantic information and designs multi‐granularity features for final relation classification. Our model offers a momentous improvement over the published methods for the relation extraction on the widely used data sets.

This paper presents an efficient graph semisupervised learning (GSSL) method that meets the criterion of optimization without iterations. Most existing GSSL methods require iterative optimization to achieve a preset objective because they consider data points to be in peer-to-peer relationships. Additionally, existing GSSL methods must learn from scratch for unseen data because graph structures are specifically built for a given dataset. By leveraging the partial order relationships induced by the local density and distances between data, we developed a novel label propagation algorithm based on the data structure of an optimal leading forest (OLeaF). The time complexity of our method is O(N) for both labeling unclassified data and labeling new data from a dataset after an OLeaF is constructed. Therefore, the two main weaknesses of traditional GSSL are addressed. Additionally, the constructed leading forest offers good interpretability for learning results. We scale the proposed method to accommodate big data by utilizing the block distance matrix technique and locality-sensitive hashing. Extensive experiments on datasets with different characteristics demonstrate the superior efficiency and competitive accuracy of the proposed method.

The problem of initialization of active learning is considered in this paper. Especially, this paper studies the problem in an imbalanced data scenario, which is called as class-imbalance active learning cold-start. The novel method is two-stage clustering-based active learning cold-start (ALCS). In the first stage, to separate the instances of minority class from that of majority class, a multi-center clustering is constructed based on a new inter-cluster tightness measure, thus the data is grouped into multiple clusters. Then, in the second stage, the initial training instances are selected from each cluster based on an adaptive candidate representative instances determination mechanism and a clusters-cyclic instance query mechanism. The comprehensive experiments demonstrate the effectiveness of the proposed method from the aspects of class coverage, classification performance, and impact on active learning.

In some cases, the decision process of three-way decisions (3WD) is costly, and sequential three-way decisions (S3WD) may cause errors beyond tolerance. To solve the above problems, in this paper, democratic three-way decisions based on voting mechanism (D3WD-VM) is proposed from the perspective of all conditional attributes. By obtaining decision opinions of different attributes at the coarse granularity level, the final decision results is obtained. First, a voting mechanism is established to realize the idea of the democratic three-way, which is an ensemble decision space based on conditional attributes. Next, in order to make the decision results more reasonable, the normalized information gain ratio is utilized to optimize the voting weight of conditional attributes in the voting mechanism. Then, based on cognitive science, two different decision strategies are devised to make the final decision. Finally, the experimental results demonstrate that the accuracy rate and the comprehensive evaluation index of the D3WD-VM have also been improved to some extent compared with the S3WD, and the decision efficiency is better than 3WD.

Recommender systems (RSs) commonly adopt a user-item rating matrix to describe users’ preferences on items. With users and items exploding, such a matrix is usually high-dimensional and sparse (HiDS). Recently, the idea of deep learning has been applied to RSs. However, current deep-structured RSs suffer from high computational complexity. Enlightened by the idea of deep forest, this paper proposes a deep latent factor model (DLFM) for building a deep-structured RS on an HiDS matrix efficiently. Its main idea is to construct a deep-structured model by sequentially connecting multiple latent factor (LF) models instead of multilayered neural networks through a nonlinear activation function. Thus, the computational complexity grows linearly with its layer count, which is easy to resolve in practice. The experimental results on four HiDS matrices from industrial RSs demonstrate that when compared with state-of-the-art LF models and deep-structured RSs, DLFM can well balance the prediction accuracy and computational efficiency, which well fits the desire of industrial RSs for fast and right recommendations.

Nonnegative matrix factorization is comprehensively used in recommendation systems. In an effort to reduce the recommended cost of newly added samples, incremental nonnegative matrix factorization and its variants have been extensively studied in recommendation systems. However, the recommendation performance is incapable of particular applications in terms of data sparsity and sample diversity. In this paper, we propose a new incremental recommend algorithm by improving incremental nonnegative matrix factorization based on three-way decision, called Three-way Decision Recommendations Based on Incremental Non-negative Matrix Factorization (3WD-INMF), in which the concept of positive, negative, and boundary regions are employed to update the new coming samples’ features. Finally, experiments on six public data sets demonstrate the error induced by 3WD-INMF is decreasing as the addition of new samples and deliver state-of-the-art performance compared with existing recommendation algorithms. The results indicate our method is more reasonable and efficient by leveraging the idea of three-way decision to perform the recommendation decision process.

Aiming to reveal the potential relationships between users, link prediction has been considered as a fundamental research issue in signed social networks. The key of the link prediction is to measure the similarity between users. Many existing researches use connections between users and their common neighbors to measure the similarities, and these methods rely too much on the structure of social networks. Most of them use the deep neural network to enhance the prediction accuracy. However, the complete structure of the huge social network cannot be captured easily, and the models learnt by the deep neural network are unexplainable and uncontrolled. As an explainable model, functional network is a recent replacement for standard neural network. Therefore, we revise the traditional strategy of functional network and propose a novel functional network framework. Firstly, the attributes are preprocessed through the cloud model to define their importance before inputting them into the functional network. Then the association algorithm is used to do aggregate computation in computing neurons for defining the connections between neurons well. Finally, we use three-way decisions to process the samples in the boundary to optimize the performance of model. Experiments executed on six real datasets show that our method has significantly higher link prediction precision than the state-of-the-art works. From our discussions, the improved functional network can be a valid replacement for neural networks in some fields.

Remote sensing scene classification is a fundamental responsibility of earth observation, aiming at identifying information granular for land cover classification. The multi-granular land use for multi-source remotely sensed image categories is now a principal task in remote sensing data augmentation and data selection. Understanding image representations are meaningful for the scene classification task. Training deep learning model-based approaches has to do with scene classification and brings about fantastic achievement. At the same time, these high-level approaches are computationally expensive and time–consuming. This paper introduces multi-granularity Neural Network Encoding architecture base on InceptionV3, InceptionReseNetV2, VGG16, and DenseNet201 architecture into remote sensing scene classification. To improve performance and to solve intra-class variation for multi-class scene issues remote sensing dataset. By using pre-trained CNN, activation function and ensemble learning have been adopting. InceptionV3 and VGG16 are used to extract features. InceptionResNetV2 use for fine-tuning, which consists of unfreezing the entre model and retraining the new data with a lower learning rate. The proposed fine-tune whole pre-trained model produces better results of test set up to 97.84 % than features extracted by InceptionResNetV2. Also, we use DCNNs ensemble average and weighted average to achieve better outcomes for 97.36% and 99.10%. In our experiments, we attempted to fine-tune the deep convolutional neural networks (DCNNs) training method for remote sensing scene classification on two public datasets UCM, SIRI-WHU, and one dataset I collected through the google earth engine from East Africa Community Countries (EACC) within nine classes within total 2112 labeled images. The results indicate that our proposed fine-tuning of the pre-trained model with few epochs and less computational time increases accuracy.

It is well known that the number of users is increasing rapidly in online social networks. People are linked through various types of social relationships. Detecting the type of social relationships is fundamental to improve performance on many applications in social networks. Existing studies mainly focus on predicting social relationships in the same network based on its own abundant information. However, there are few works on predicting social relationships across different networks. In this paper, we study a tri-training model to predict the labels of social relationships across different networks based on BP neural networks. Firstly, we aim to obtain the hidden common characteristics between two different networks. Depending on these hidden common characteristics, unlabeled target samples are assigned pseudo labels by training two BP neural networks. Then we capture the special features for the target network by analyzing the structural characters. Finally, combining with the special and common features, we optimize the third BP neural network by training the pseudo labeled target samples. After training, we can use the third optimized BP neural network to predict the types of social relationships in the target network. We evaluate the model and compare it with some existing methods on six real online social networks. The experimental results indicate that our model can effectively exploit the unlabeled target sample and outperform the existing methods on multiple metrics, and also show that the special features in the target network can help to enhance the learning performance.

Reinforcement learning (RL) has been widely used to aid training in language generation. This is achieved by enhancing standard maximum likelihood objectives with user-specified reward functions that encourage global semantic consistency. We propose a principled approach to address the difficulties associated with RL-based solutions, namely, high-variance gradients, uninformative rewards and brittle training. By leveraging the optimal transport distance, we introduce a regularizer that significantly alleviates the above issues. Our formulation emphasizes the preservation of semantic features, enabling end-to-end training instead of ad-hoc fine-tuning, and when combined with RL, it controls the exploration space for more efficient model updates. To validate the effectiveness of the proposed solution, we perform a comprehensive evaluation covering a wide variety of NLP tasks: machine translation, abstractive text summarization and image caption, with consistent improvements over competing solutions.

In recent years, multi-view clustering algorithms have shown promising performance by combining multiple sources or views of datasets. A problem that has not been addressed satisfactorily is the uncertain relationship between an object and a cluster. Thus, this paper investigates an active three-way clustering method via low-rank matrices that can improve clustering accuracy as clustering proceeds for the multi-view data of high dimensionality. We adopt a three-way clustering representation to reflect the three types of relationships between an object and a cluster, namely, belong-to definitely, uncertain and not belong-to definitely. We construct the consensus low-rank matrix from each weighted low-rank matrix by taking account of the diversity of views, and give the method to solve the optimization problem of objective function based on the improved augmented Lagrangian multiplier algorithm. We suggest an active learning strategy to learn important informative pairwise constraints after measuring the uncertainty of an object based on the entropy concept. The experimental results conducted on real-world datasets have validated the effectiveness of the proposed method.

Cluster ensemble has emerged as a powerful technique for combining multiple clustering results. To address the problem of clustering on large-scale data, this paper presents an efficient three-way cluster ensemble approach based on Spark, which has the ability to deal with both hard clustering and soft clustering. First, this paper proposes the framework of three-way cluster ensemble based on Spark inspired by the theory of three-way decisions, and develops a distributed three-way k-means clustering algorithm. Then, we introduce the concept of cluster unit, which reflects the minimal granularity distribution structure agreed by all the ensemble members. We also introduce quantitative measures for calculating the relationships between units and between clusters. Finally, we propose a consensus clustering algorithm based on cluster units, and we devise various three-way decision strategies to assign small cluster units and no-unit objects. The experimental results using 19 real-world data sets validate the effectiveness of the proposed approach from different indices such as ARI, ACC, NMI and F1-Measure. The experimental results show that the proposed approach can effectively deal with large-scale data, and the proposed consensus clustering algorithm has a lower time cost and does not sacrifice the clustering quality.

The existing noise detection methods required the classifiers or distance measurements or data overall distribution, and ‘curse of dimensionality’ and other restrictions made them insufficiently effective in complex data, e.g., different attribute weights, high-dimensionality, containing feature noise, nonlinearity, etc. This is also the main reason that the existing noise filtering methods were not widely applied and formed an effective learning framework. To address this problem, we propose here a complete and efficient random forest method (CRF) specifically for the class noise detection by simulating the grid generation and expansion. The CRF is not based on distance measures or overall distribution or classifiers; besides, the voting mechanism makes it able to effectively process datasets containing feature noise. Furthermore, we introduce CRF based class noise filtering learning framework (CRF-NFL) and derive its mathematical model. The framework is then applied to many widely used classifiers including some state-of-the-art algorithms, e.g., k-means tree, GBDT, and XGBoost. Moreover, its parallelized is designed for large-scale data. The CRF-NFL show much better generalizability than the conventional classifiers and the relative density-based method, which is the most effective noise filtering method as far as we know. All research has formed an open source library, called CRF-NFL: http://www.cquptshuyinxia.com/CRF-NFL.html.

The existing multi-view clustering algorithms require that a sample in a view is completely or partially mapped onto one or more samples in a different corresponding view. However, this requirement could not be satisfied in many practical applications. Fortunately, there is a common cognition that the graph structure formed from each view should be as consistent as possible. Thus, this paper proposes a novel multi-view clustering method for unknown mapping relationships between cross-view samples based on the framework of non-negative matrix factorization, as an attempt to solve this problem. The objective function is designed by effectively building reconstruction error terms, local structural constraint terms, and cross-view mapping loss terms by exploring cross-view relationships. The experimental results show that the proposed method not only performs well to reveal the real mapping relationships between cross-view samples but also outperforms the comparison algorithms on the obtained clustering results.

There are three types of relationships between an object and a cluster, namely, belong-to definitely, uncertain and not belong-to definitely. Most of the existing clustering algorithms represent a cluster with a single set and they are the two-way clustering algorithms since they just reflect two relationships. By contrast, the three-way clustering can reflect intuitively the three types of relationships with a pair of sets. However, the three-way clustering algorithms usually need to know the thresholds in advance in order to obtain the three types of relationships. To address the problem, we propose an efficient three-way clustering algorithm based on the idea of universal gravitation in this paper. The proposed method can adjust the thresholds automatically in the process of clustering and obtain more detailed ascription relation between objects and clusters. Furthermore, to guarantee the integrity of the work, we also put forward a two-way clustering algorithm to obtain the conventional two-way result. The experimental results show that the proposed algorithm is not only effective to obtain the three-way clustering result from the two-way clustering result automatically, but also it is in a better performance at the accuracy, F-measure, NMI and RI than the compared algorithms in most cases.

Natural Language Generation (NLG) is a key component in a task-oriented dialogue system, which converts the structured meaning representation (MR) to the natural language. For large-scale conversational systems, where it is common to have over hundreds of intents and thousands of slots, neither template-based approaches nor model-based approaches are scalable. Recently, neural NLGs started leveraging transfer learning and showed promising results in few-shot settings. This paper proposes AUGNLG, a novel data augmentation approach that combines a self-trained neural retrieval model with a few-shot learned NLU model, to automatically create MR-to-Text data from open-domain texts. The proposed system mostly outperforms the state-of-the-art methods on the FewShotWOZ data in both BLEU and Slot Error Rate. We further confirm improved results on the FewShotSGD data and provide comprehensive analysis results on key components of our system. Our code and data are available at https://github.com/XinnuoXu/AugNLG.

The problem of two-sided (bilateral) matching is a process of matching demanders and suppliers reasonably in supply chain. The vast majority of existing researches are based on a single-stage static evaluation information. It is necessary to consider the dynamic evaluation information during multiple stages since the performances of both matching parties have some certain dynamic characteristics. To address the problem, this paper proposes a two-sided matching method based on multi-stage dynamic reference points, because the decision-makers often focus on the level of gain or loss of the evaluation object relative to a reference point in the view of prospect theory.

Neural language models are often trained with maximum likelihood estimation (MLE), where the next word is generated conditioned on the ground-truth word tokens. During testing, however, the model is instead conditioned on previously generated tokens, resulting in what is termed exposure bias. To reduce this gap between training and testing, we propose using optimal transport (OT) to match the sequences generated in these two modes. An extension is further proposed to improve the OT learning, based on the structural and contextual information of the text sequences. The effectiveness of the proposed method is validated on machine translation, text summarization, and text generation tasks., To appear at EMNLP 2020

Relation classification is a very important Natural Language Processing (NLP) task to classify the relations from the plain text. It is one of the basic tasks of constructing a knowledge graph. Most existing state-of-the-art methods are primarily based on Convolutional Neural Networks(CNN) or Long Short-Term Memory Networks(LSTM). Recently, many pre-trained Bidirectional Encoder Representation from Transformers (BERT) models have been successfully used in the sequence labeling and many NLP classification tasks. Relation classification is different in that it needs to pay attention to not only the sentence information but also the entity pairs. In this paper, a Siamese BERT model with Adversarial Training (SBERT-AT) is proposed for relation classification. Firstly, the features of the entities and the sentence can be extracted separately to improve the performance of relation classification. Secondly, the adversarial training is applied to the SBERT architecture to improve the robustness. Lastly, the experimental results demonstrate that we achieve significant improvement compared with the other methods on real-world datasets.

Auto-regressive text generation models usually focus on local fluency, and may cause inconsistent semantic meaning in long text generation. Further, automatically generating words with similar semantics is challenging, and hand-crafted linguistic rules are difficult to apply. We consider a text planning scheme and present a model-based imitation-learning approach to alleviate the aforementioned issues. Specifically, we propose a novel guider network to focus on the generative process over a longer horizon, which can assist next-word prediction and provide intermediate rewards for generator optimization. Extensive experiments demonstrate that the proposed method leads to improved performance., ACL 2020. arXiv admin note: substantial text overlap with arXiv:1811.00696

A spatial co-location pattern represents a subset of spatial features whose instances are prevalently located together in a geographic space. Although many algorithms of mining spatial co-location pattern have been proposed, there are still some problems: 1) they miss some meaningful patterns (e.g., {Ganoderma_lucidumnew, maple_treedead} and {water_hyacinthnew(increase), algaedead(decrease)}), and get the wrong conclusion that the instances of two or more features increase/decrease (i.e., new/dead) in the same/approximate proportion, which has no effect on prevalent patterns. 2) Since the number of prevalent spatial co-location patterns is very large, the efficiency of existing methods is very low to mine prevalent spatial co-location patterns. Therefore, first, we propose the concept of dynamic spatial co-location pattern that can reflect the dynamic relationships among spatial features. Second, we mine small number of prevalent maximal dynamic spatial co-location patterns which can derive all prevalent dynamic spatial co-location patterns, which can improve the efficiency of obtaining all prevalent dynamic spatial co-location patterns. Third, we propose an algorithm for mining prevalent maximal dynamic spatial co-location patterns and two pruning strategies. Finally, the effectiveness and efficiency of the method proposed as well as the pruning strategies are verified by extensive experiments over real/synthetic datasets., Comment: 10 pages,7 figures

Pretrained Language Models (PLMs) have improved the performance of natural language understanding in recent years. Such models are pretrained on large corpora, which encode the general prior knowledge of natural languages but are agnostic to information characteristic of downstream tasks. This often results in overfitting when fine-tuned with low resource datasets where task-specific information is limited. In this paper, we integrate label information as a task-specific prior into the self-attention component of pretrained BERT models. Experiments on several benchmarks and real-word datasets suggest that the proposed approach can largely improve the performance of pretrained models when fine-tuning with small datasets.

We propose a novel graph-driven generative model, that unifies multiple heterogeneous learning tasks into the same framework. The proposed model is based on the fact that heterogeneous learning tasks, which correspond to different generative processes, often rely on data with a shared graph structure. Accordingly, our model combines a graph convolutional network (GCN) with multiple variational autoencoders, thus embedding the nodes of the graph i.e., samples for the tasks) in a uniform manner while specializing their organization and usage to different tasks. With a focus on healthcare applications (tasks), including clinical topic modeling, procedure recommendation and admission-type prediction, we demonstrate that our method successfully leverages information across different tasks, boosting performance in all tasks and outperforming existing state-of-the-art approaches., Accepted by AAAI-2020

Self-labeled technique, a paradigm of semisupervised classification (SSC), is highly effective in alleviating the shortage of labeled data in classification tasks via an iterative self-labeling process. Although existing self-labeled SSC models show great prospect in industrial applications, they suffer from performance degeneration caused by false-positive label-predictions of unlabeled data during the iterative self-labeling process. For addressing this issue, this paper proposes a novel SSC framework, which is highly compatible with most existing self-labeled SSC models. The main idea of this framework is to incorporate a differential-evolution-based positioning optimization algorithm for classification into the iterative self-labeling process, aiming at optimizing the positioning of newly labeled data. Specifically, five representative self-labeled SSC models with different characteristics are modified based on the proposed framework to check their performances. Experimental results on 45 benchmark datasets demonstrate that the proposed framework is highly compatible with tested self-labeled SSC models, and significantly effective in improving their performances.

The three-way decision model is a topic of substantial research interest in the field of artificial intelligence, and many researchers have focused on to its feasibility and rationality. The tolerance and practicability of the three-way decision model are better than those of the two-way decision model. When the attribute value of each object in a domain is given, the formation of a three-way classification of the domain is a key issue. However, few studies have been conducted on establishing a three-way decision model with the given attribute values in the case where the number of objects in an accepted region is given. Therefore, in the model presented in this paper, both the uncertainty of attribute values and the cost of updating are fully considered. In this paper, first, a new concept of attribute ratio is defined to describe an object when the attribute value of the object is numerical, and then, a dynamic three-way decision model is established. Second, a feature extraction algorithm of attribute values is proposed, and a pair of decision thresholds of the dynamic three-way decision model is also obtained according to the given conditions. Then, in the case where the attribute values are updated, an example is provided to demonstrate how two-way classification results can be obtained in the dynamic decision-making process. Finally, the results of simulation experiments show that the proposed model is feasible and effective in practical applications. When the number of objects in an accepted region has been given, according to the updating strategy of attribute values, the three-way decision problems are successfully solved by the proposed model.

Having a multitude of unlabeled data and few labeled ones is a common problem in many practical applications. A successful methodology to tackle this problem is self-training semi-supervised classification. In this paper, we introduce a method to discover the structure of data space based on find of density peaks. Then, a framework for self-training semi-supervised classification, in which the structure of data space is integrated into the self-training iterative process to help train a better classifier, is proposed. A series of experiments on both artificial and real datasets are run to evaluate the performance of our proposed framework. Experimental results clearly demonstrate that our proposed framework has better performance than some previous works in general on both artificial and real datasets, especially when the distribution of data is non-spherical. Besides, we also find that the support vector machine is particularly suitable for our proposed framework to play the role of base classifier.

For an uncertain target set, three disjoint regions, i.e., positive, negative, and boundary regions, are established in a rough approximation space. To improve the fault-tolerance ability of the rough set model, a probabilistic rough set model with a pair of thresholds is proposed, and two types of classification errors are introduced from the viewpoint of probability and statistics. A decision-theoretic rough set model is proposed to minimize the decision risk, and a pair of thresholds of the probabilistic rough set model can be calculated by the given cost parameters. In this paper, first, two types of classification errors and two types of uncertain classifications are defined in the probabilistic rough set model. Then, by considering four cost parameters that characterize the cost of two types of classification errors and two types of uncertain classifications, a pair of thresholds of the probabilistic rough set model can again be obtained. Finally, with the increase in new attributes, the change regularities of the total cost with changing rough approximation spaces are discussed in detail, and experimental results show that the proposed model for three-way decision is reasonable and effective. These results further enrich three-way decision theory to effectively deal with uncertain classification problems.

The probabilistic rough set model is a generalized rough set model, which was proposed to improve the fault tolerance ability of the classical Pawlak’s rough set model. Compared with the classical Pawlak’s rough set model, the probabilistic rough set model and its several special forms (such as variable precision rough set model, Bayesian rough set model, and decision-theoretic rough set model) divide a domain into three regions by two parameters ( ${\alpha , \beta }$ ), and the objects belong to every region with a certain uncertainty. With the increase of information (attributes), the boundary region of the probabilistic rough set model may become smaller, bigger, or remain unchanged. This paper aims first to study some more complex uncertainty of the probabilistic rough set model. Then the uncertainty of the three regions is defined and analyzed. Besides, the changing regularities of uncertainty of a target concept in the probabilistic rough set model are discovered. Finally, three kinds of incremental information are defined, and their judging theorems are proposed. These results could enrich and improve rough set theory to deal with uncertain information systems.

This investigation developed a hybrid rough set (RST) and multidimensional cloud model (RSMCM) to leverage the unique strengths of RST and cloud modeling to evaluate the trophic level. In the proposed hybrid model, RST is used to decrease the data scale and extract the qualitative rules, and the multidimensional cloud model is employed to quantitatively analyze the average values, uniformity and stability of water eutrophication. The experimental results reveal that the hybrid model achieves more accurate assessment results than other mainstream models. Therefore, the hybrid model is a promising alternative for a water eutrophication information system and offers a quantitative measure for evaluating the uniformity and stability of eutrophication in utilities management and for operations staff.

Detecting clusters of arbitrary shape and constantly delivering the results for newly arrived items are two critical challenges in the study of data stream clustering. However, the existing clustering methods could not deal with these two problems simultaneously. In this paper, we employ the density peaks based clustering (DPClust) algorithm to construct a leading tree (LT) and further transform it into a fat node leading tree (FNLT) in a granular computing way. FNLT is a novel interpretable synopsis of the current state of data stream for clustering. New incoming data is blended into the evolving FNLT structure quickly, and thus the clustering result of the incoming data can be delivered on the fly. During the interval between the delivery of the clustering results and the arrival of new data, the FNLT with blended data is granulated as a new FNLT with a constant number of fat nodes. The FNLT of the current data stream is maintained in a real-time fashion by the Blending-Granulating-Fading mechanism. At the same time, the change points are detected using the partial order relation between each pair of the cluster centers and the martingale theory. Compared to several state-of-the-art clustering methods, the presented model shows promising accuracy and efficiency.

Many high-level dimensionality reduction approaches for mining time series have been proposed, e.g., SAX, PWCA , and Feature-based. Due to the rapid performance degradation of time-series data mining in much lower dimensionality and the continuously increasing amount of time series data with uncertainty, there remains a burning need to develop new time-series representations that can retain good performance in much lower reduced space and address uncertainty efficiently. In this work, we propose a novel time series representation, namely Two-dimensional Normal Cloud Representation (2D-NCR), based on cloud model theory. The representation achieves dimensionality reduction by transforming the raw time series into a sequence of two-dimensional normal cloud models. Moreover, a new similarity measure between the transformed time series is presented. The proposed method can reflect the characteristic data distribution of the time series and capture the variation with time. We validate the performance of our representation on the various data mining tasks of classification, clustering, and query by content. The experimental results demonstrate that 2D-NCR is an effective and competitive representation for time-series data mining.

The three-way representation of communities as a triple of positive, boundary and negative regions brings more insight into interpretation of communities.The overlapping regions are further refined into 4 macro types and 8 micro types.The three-way decision strategies are introduced to detect the overlapping communities based on the introduced relation-graph.A new fitness function is devised to adapt to processing networks with overlapping.The refinement of overlapping regions can reveal the different significance to impact forming or changing of communities. The identification of communities is crucial to an understanding of the structural and functional properties of a complex network. Many methods and algorithms have been developed to detect overlapping communities. A problem that has not been addressed satisfactorily is the relationship or difference between vertices in overlapping regions during the formation and growth of communities. This paper investigates methods that not only detect the overlapping communities but also refine the overlapping regions. We give a three-way representation of a community by using interval sets and re-formalize the problem of community detection as three-way clustering. We suggest four macro types and eight micro types of vertices to characterize members in overlapping regions for their refinement. We propose an overlapping community detection algorithm by classifying the vertices into core vertices, bone vertices, and trivial vertices. The main strategy of this algorithm is to find an initial cluster core, to expand the core to a preliminary community according to a new fitness function, and to merge trivial vertices based on three-way decision strategies. The experimental results on both real-world social networks and computer-generated artificial networks show the effectiveness and efficiency of the proposed methods.

Existing hierarchical clustering algorithms involve a flat clustering component and an additional agglomerative or divisive procedure. This paper presents a density peak based hierarchical clustering method (DenPEHC), which directly generates clusters on each possible clustering layer, and introduces a grid granulation framework to enable DenPEHC to cluster large-scale and high-dimensional (LSHD) datasets. This study consists of three parts: (1) utilizing the distribution of the parameter γ , which is defined as the product of the local density ρ and the minimal distance to data points with higher density δ in “clustering by fast search and find of density peaks” (DPClust), and a linear fitting approach to select clustering centers with the clustering hierarchy decided by finding the “stairs” in the γ curve; (2) analyzing the leading tree (in which each node except the root is led by its parent to join the same cluster) as an intermediate result of DPClust, and constructing the clustering hierarchy efficiently based on the tree; and (3) designing a framework to enable DenPEHC to cluster LSHD datasets when a large number of attributes can be grouped by their semantics. The proposed method builds the clustering hierarchy by simply disconnecting the center points from their parents with a linear computational complexity O ( m ), where m is the number of clusters. Experiments on synthetic and real datasets show that the proposed method has promising efficiency, accuracy and robustness compared to state-of-the-art methods.

After probability theory, fuzzy set theory and evidence theory, rough set theory is a new mathematical tool for dealing with vague, imprecise, inconsistent and uncertain knowledge. In recent years, the research and applications on rough set theory have attracted more and more researchers' attention. And it is one of the hot issues in the artificial intelligence field. In this paper, the basic concepts, operations and characteristics on the rough set theory are introduced firstly, and then the extensions of rough set model, the situation of their applications, some application software and the key problems in applied research for the rough set theory are presented.

Relation classification is an important semantic processing task in natural language processing (NLP). Traditional works on relation classification are primarily based on supervised methods and distant supervision which rely on the large number of labels. However, these existing methods inevitably suffer from wrong labeling problem and may not perform well in resource-poor domains. We thus utilize transfer learning methods on relation classification to enable relation classification system to adapt resource-poor domains along with different relation type. In this paper, we exploit a convolutional neural network to extract lexical and syntactic features and apply transfer learning approaches for transferring the parameters of convolutional layer pre-training on general-domain corpus. The experimental results on real-world datasets demonstrate that our approach is effective and outperforms several competitive baseline methods.

It is a basic task to measure the similarity between two uncertain concepts in many real-life artificial intelligence applications, such as image retrieval, collaborative filtering, public opinion guidance, and so on. As an important cognitive computing model, cloud model has been used in many fields of artificial intelligence. It can realise the bidirectional cognitive transformation between qualitative concept and quantitative data based on the theory of probability and fuzzy set. The similarity measure of two uncertain concepts is a fundamental issue in cloud model theory. Popular similarity measure methods of cloud model are surveyed in this study. Their limitations are analysed in detail. Some related future research topics are proposed.

Deep neural networks can obtain effective hierarchical representations, which make them perform well in image recognition and speech recognition. However, the disadvantages of deep learning relying on large-scale annotation data also limit its development. To this end, a multi-layer convolutional neural network based on knowledge graph prior knowledge is proposed in this paper, which acquires text features from finer granularity, and using the existing prior knowledge which increases granularity information to enhance the semantic information of the text, and to reduce the dependence of the model on large-scale samples. At the same time, it can reduce the over-fitting problem of the model, and improve classification accuracy. This model performs well in several large data sets. By introducing prior knowledge to TextCNN, a classical deep neural model, it has also verified the validity of using prior knowledge through experiments.

Accurately predicting unknown quality-of-service (QoS) data based on historical QoS records is vital in web service recommendation or selection. Recently, latent factor (LF) model has been widely and successfully applied to QoS prediction because it is accurate and scalable under many circumstances. Hence, state-of-the-art methods in QoS prediction are primarily based on LF. They improve the basic LF-based models by identifying the neighborhoods of QoS data based on some additional geographical information. However, the additional geographical information may be difficult to collect in considering information security, identity privacy, and commercial interests in real-world applications. Besides, they ignore the reliability of QoS data while unreliable ones are often mixed in. To address these issues, this paper proposes a data-aware latent factor (DALF) model to achieve highly accurate QoS prediction, where ‘data-aware’ means DALF can easily implement the predictions according to the characteristics of QoS data. The main idea is to incorporate a density peaks based clustering method into an LF model to discover the neighborhoods and unreliable ones of QoS data. Experimental results on two benchmark real-world web service QoS datasets demonstrate that DALF has better performance than the state-of-the-art models.

Extracting entities and their relations expressed in free text is essential to correct and populate knowledge graphs. Traditional methods assume that only the information of entities benefits the extraction of relations. They view this task as a two-step task, named entity recognition (NER) and relation classification (RC). However, the inadequate use of information and the error propagation problem constrain methods following this pipeline fashion. Joint extraction methods are proposed to incorporate useful interaction information between the two tasks for improvement, which solve NER and RC simultaneously. Although they have been proved to be superior to pipeline models, their performance is still far from satisfaction. In this paper, we try to combine the idea of data-driven granular cognitive computing and deep learning in joint extraction task. Accordingly, a neural-based joint extraction model named Joint extraction with Multi-granularity Context (JMC) is proposed. It explores the multi-granularity context of natural language sentences and uses neural networks to learn representations of these context automatically. Experiments results on NYT, a large data set produced by the distant supervision technique, show that JMC achieves comparative results to state-of-the-art methods.

Generating high-quality paraphrases is a fundamental yet challenging natural language processing task. Despite the effectiveness of previous work based on generative models, there remain problems with exposure bias in recurrent neural networks, and often a failure to generate realistic sentences. To overcome these challenges, we propose the first end-to-end conditional generative architecture for generating paraphrases via adversarial training, which does not depend on extra linguistic information. Extensive experiments on four public datasets demonstrate the proposed method achieves state-of-the-art results, outperforming previous generative architectures on both automatic metrics (BLEU, METEOR, and TER) and human evaluations.

A hierarchical organization is common in most corporations, where some entities (objects) in the lower granulation level are subordinate to a single other entity (object) in the follow-on higher granulation level. During recent years, how to evaluate the objects in a group corporation has become a key strategic consideration. Therefore, this paper proposes a multi-granular hierarchical evaluation model for multiple criteria three sorting. In the model, the objects in a granulation level are assigned to three ordinal classes based on the theory of three-way decisions and the rating system, instead of accurately sorting them. Furthermore, the evaluation results from the previous granulation level are employed as one type of criteria in the follow-on granulation level, called the evaluation result criteria. Additionally, we propose a three-way TOPSIS (the Technique for Order of Preference by Similarity to Ideal Solution) method based on the characteristics of four types of criteria such as evaluation result criteria, benefit criteria, cost criteria and non-monotonic criteria. Besides, a pre-evaluation strategy is devised at each level inspired by the concept of "first impressions". Then, the final evaluation results are obtained by studying the qualitative evaluation results of pre-evaluation method and the quantitative evaluation results of three-way TOPSIS method. Finally, a numerical case study from an aluminium electrolysis production enterprise is illustrated to show its viability.

Community structure detection is an important and valuable task in social network studies as it is the base for many social network applications such as link prediction, recommendation, etc. Most social networks have an inherent multi-granular structure, which leads to different community structures at different granularities. However, few studies pay attention to such multi-granular characteristics of social networks. In this paper, a method called MGCD (Multi-Granularity Community Detection) is proposed for finding multi-granularity community structures of social networks. At first, a network embedding method is used to obtain the low-dimensional vector representation for each node. Then, an effective embedding-based strategy for weakening the detected community structures is proposed. Finally, a joint learning framework, which combines network embedding and community structure weakening is developed for identifying the multi-granularity community structures of social networks. Experimental results on real-world networks show that MGCD outperforms the state-of-the-art benchmark methods on finding multi-granularity community structure tasks.

Rough set provides a theoretical framework for classification learning in data mining and knowledge discovery. As an important application of rough set, attribute reduction, also called feature selection, aims to reduce the redundant attributes in a given decision system while preserving a particular classification property, e.g., information entropy and knowledge granularity. In view of the dynamic changes of the object set in a decision system, in this paper, we focus on knowledge granularity-based attribute reduction approach when some objects vary dynamically. We first introduce incremental mechanisms to compute new knowledge granularity. Then, the corresponding incremental algorithms for attribute reduction are developed when some objects are added into and deleted from the decision system. Experiments conducted on different data sets from UCI show that the proposed incremental algorithm can achieve better performance than the non-incremental counterpart and incremental algorithm based on entropy.

In "traditional" rough set methodologies, data are assumed to be stored in a single data repository. However, this assumption is not always true in many real-world problems, where data may be distributed across multiple locations, which is especially pertinent with the development of the Internet. To cope with this phenomenon, in this paper we extend the methodology of rough sets to distributed decision information systems. We first present a definition of rough sets in distributed decision information systems. Then we study the reducibility of distributed decision information systems at two different levels of granularity. The conditions for a decision information table or an attribute in distributed decision information systems to be reducible are presented, and an approach to compute reducts of a distributed decision information system is developed. The experimental results show that the proposed approach can be used to simplify distributed decision information systems, while retain their classification abilities.

Incomplete contexts are a kind of formal contexts in which the relationship between some objects and some attributes is unavailable or lost. Knowledge discovery in incomplete contexts is of interest because such databases are frequently encountered in the real world. This paper mainly focuses on two issues: approximate concept construction with three-way decisions and attribute reduction in incomplete contexts. The theory of three-way decisions is formulated based on the notions of acceptance, rejection and non-commitment. It is an extension of the commonly used binary-decision model with an added third option. Based on three-way decisions, we propose two models to construct approximate concepts in incomplete contexts, and the equivalence of the two is revealed. To simplify the representation of the approximate concept lattices, we further present the attribute reduction approaches.

In spite of the impressive diversity of multi-factors fuzzy time series models, there is still a burning need to develop models that can handle the uncertainty inherent in certain data with some certain methods and obtain high forecasting accuracy. This paper proposes a novel multi-granularity combined prediction model for multi-factors fuzzy time series forecasting. The proposed model utilizes the clustering algorithm to generate different lengths of intervals, and forecasts the fuzzy trend in different granular spaces. It calculates the final forecasted value by using the forecasted fuzzy trend in each granular space and the optimal weighting vector obtained by particle swarm techniques. The proposed model can transform the uncertain problem to a certain problem through the granular computing theory. The experiments show that the presented model can not only obtain higher forecasting accuracy than several existing methods to forecast the Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) and the civilian unemployment rate, but also capture and interpret the fuzzy trend.